4 * Copyright (C) 1991, 1992 Linus Torvalds
7 #include <linux/syscalls.h>
8 #include <linux/init.h>
11 #include <linux/file.h>
12 #include <linux/capability.h>
13 #include <linux/dnotify.h>
14 #include <linux/smp_lock.h>
15 #include <linux/slab.h>
16 #include <linux/module.h>
17 #include <linux/security.h>
18 #include <linux/ptrace.h>
19 #include <linux/signal.h>
20 #include <linux/rcupdate.h>
23 #include <asm/siginfo.h>
24 #include <asm/uaccess.h>
26 void fastcall set_close_on_exec(unsigned int fd, int flag)
28 struct files_struct *files = current->files;
30 spin_lock(&files->file_lock);
31 fdt = files_fdtable(files);
33 FD_SET(fd, fdt->close_on_exec);
35 FD_CLR(fd, fdt->close_on_exec);
36 spin_unlock(&files->file_lock);
39 static inline int get_close_on_exec(unsigned int fd)
41 struct files_struct *files = current->files;
45 fdt = files_fdtable(files);
46 res = FD_ISSET(fd, fdt->close_on_exec);
52 * locate_fd finds a free file descriptor in the open_fds fdset,
53 * expanding the fd arrays if necessary. Must be called with the
54 * file_lock held for write.
57 static int locate_fd(struct files_struct *files,
58 struct file *file, unsigned int orig_start)
66 if (orig_start >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
70 fdt = files_fdtable(files);
72 * Someone might have closed fd's in the range
73 * orig_start..fdt->next_fd
76 if (start < fdt->next_fd)
80 if (start < fdt->max_fdset) {
81 newfd = find_next_zero_bit(fdt->open_fds->fds_bits,
82 fdt->max_fdset, start);
86 if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
89 error = expand_files(files, newfd);
94 * If we needed to expand the fs array we
95 * might have blocked - try again.
101 * We reacquired files_lock, so we are safe as long as
102 * we reacquire the fdtable pointer and use it while holding
103 * the lock, no one can free it during that time.
105 fdt = files_fdtable(files);
106 if (start <= fdt->next_fd)
107 fdt->next_fd = newfd + 1;
115 static int dupfd(struct file *file, unsigned int start)
117 struct files_struct * files = current->files;
121 spin_lock(&files->file_lock);
122 fd = locate_fd(files, file, start);
124 /* locate_fd() may have expanded fdtable, load the ptr */
125 fdt = files_fdtable(files);
126 FD_SET(fd, fdt->open_fds);
127 FD_CLR(fd, fdt->close_on_exec);
128 spin_unlock(&files->file_lock);
129 fd_install(fd, file);
131 spin_unlock(&files->file_lock);
138 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
141 struct file * file, *tofree;
142 struct files_struct * files = current->files;
145 spin_lock(&files->file_lock);
146 if (!(file = fcheck(oldfd)))
152 if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
154 get_file(file); /* We are now finished with oldfd */
156 err = expand_files(files, newfd);
160 /* To avoid races with open() and dup(), we will mark the fd as
161 * in-use in the open-file bitmap throughout the entire dup2()
162 * process. This is quite safe: do_close() uses the fd array
163 * entry, not the bitmap, to decide what work needs to be
165 /* Doesn't work. open() might be there first. --AV */
167 /* Yes. It's a race. In user space. Nothing sane to do */
169 fdt = files_fdtable(files);
170 tofree = fdt->fd[newfd];
171 if (!tofree && FD_ISSET(newfd, fdt->open_fds))
174 rcu_assign_pointer(fdt->fd[newfd], file);
175 FD_SET(newfd, fdt->open_fds);
176 FD_CLR(newfd, fdt->close_on_exec);
177 spin_unlock(&files->file_lock);
180 filp_close(tofree, files);
185 spin_unlock(&files->file_lock);
189 spin_unlock(&files->file_lock);
194 asmlinkage long sys_dup(unsigned int fildes)
197 struct file * file = fget(fildes);
200 ret = dupfd(file, 0);
204 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
206 static int setfl(int fd, struct file * filp, unsigned long arg)
208 struct inode * inode = filp->f_dentry->d_inode;
211 /* O_APPEND cannot be cleared if the file is marked as append-only */
212 if (!(arg & O_APPEND) && IS_APPEND(inode))
215 /* O_NOATIME can only be set by the owner or superuser */
216 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
217 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
220 /* required for strict SunOS emulation */
221 if (O_NONBLOCK != O_NDELAY)
225 if (arg & O_DIRECT) {
226 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
227 !filp->f_mapping->a_ops->direct_IO)
231 if (filp->f_op && filp->f_op->check_flags)
232 error = filp->f_op->check_flags(arg);
237 if ((arg ^ filp->f_flags) & FASYNC) {
238 if (filp->f_op && filp->f_op->fasync) {
239 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
245 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
251 static void f_modown(struct file *filp, unsigned long pid,
252 uid_t uid, uid_t euid, int force)
254 write_lock_irq(&filp->f_owner.lock);
255 if (force || !filp->f_owner.pid) {
256 filp->f_owner.pid = pid;
257 filp->f_owner.uid = uid;
258 filp->f_owner.euid = euid;
260 write_unlock_irq(&filp->f_owner.lock);
263 int f_setown(struct file *filp, unsigned long arg, int force)
267 err = security_file_set_fowner(filp);
271 f_modown(filp, arg, current->uid, current->euid, force);
275 EXPORT_SYMBOL(f_setown);
277 void f_delown(struct file *filp)
279 f_modown(filp, 0, 0, 0, 1);
282 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
290 err = dupfd(filp, arg);
293 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
297 set_close_on_exec(fd, arg & FD_CLOEXEC);
303 err = setfl(fd, filp, arg);
306 err = fcntl_getlk(filp, (struct flock __user *) arg);
310 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
314 * XXX If f_owner is a process group, the
315 * negative return value will get converted
316 * into an error. Oops. If we keep the
317 * current syscall conventions, the only way
318 * to fix this will be in libc.
320 err = filp->f_owner.pid;
321 force_successful_syscall_return();
324 err = f_setown(filp, arg, 1);
327 err = filp->f_owner.signum;
330 /* arg == 0 restores default behaviour. */
331 if (!valid_signal(arg)) {
335 filp->f_owner.signum = arg;
338 err = fcntl_getlease(filp);
341 err = fcntl_setlease(fd, filp, arg);
344 err = fcntl_dirnotify(fd, filp, arg);
352 asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
361 err = security_file_fcntl(filp, cmd, arg);
367 err = do_fcntl(fd, cmd, arg, filp);
374 #if BITS_PER_LONG == 32
375 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
385 err = security_file_fcntl(filp, cmd, arg);
394 err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
398 err = fcntl_setlk64(fd, filp, cmd,
399 (struct flock64 __user *) arg);
402 err = do_fcntl(fd, cmd, arg, filp);
411 /* Table to convert sigio signal codes into poll band bitmaps */
413 static long band_table[NSIGPOLL] = {
414 POLLIN | POLLRDNORM, /* POLL_IN */
415 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
416 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
417 POLLERR, /* POLL_ERR */
418 POLLPRI | POLLRDBAND, /* POLL_PRI */
419 POLLHUP | POLLERR /* POLL_HUP */
422 static inline int sigio_perm(struct task_struct *p,
423 struct fown_struct *fown, int sig)
425 return (((fown->euid == 0) ||
426 (fown->euid == p->suid) || (fown->euid == p->uid) ||
427 (fown->uid == p->suid) || (fown->uid == p->uid)) &&
428 !security_file_send_sigiotask(p, fown, sig));
431 static void send_sigio_to_task(struct task_struct *p,
432 struct fown_struct *fown,
436 if (!sigio_perm(p, fown, fown->signum))
439 switch (fown->signum) {
442 /* Queue a rt signal with the appropriate fd as its
443 value. We use SI_SIGIO as the source, not
444 SI_KERNEL, since kernel signals always get
445 delivered even if we can't queue. Failure to
446 queue in this case _should_ be reported; we fall
447 back to SIGIO in that case. --sct */
448 si.si_signo = fown->signum;
451 /* Make sure we are called with one of the POLL_*
452 reasons, otherwise we could leak kernel stack into
454 if ((reason & __SI_MASK) != __SI_POLL)
456 if (reason - POLL_IN >= NSIGPOLL)
459 si.si_band = band_table[reason - POLL_IN];
461 if (!group_send_sig_info(fown->signum, &si, p))
463 /* fall-through: fall back on the old plain SIGIO signal */
465 group_send_sig_info(SIGIO, SEND_SIG_PRIV, p);
469 void send_sigio(struct fown_struct *fown, int fd, int band)
471 struct task_struct *p;
474 read_lock(&fown->lock);
477 goto out_unlock_fown;
479 read_lock(&tasklist_lock);
481 p = find_task_by_pid(pid);
483 send_sigio_to_task(p, fown, fd, band);
486 do_each_task_pid(-pid, PIDTYPE_PGID, p) {
487 send_sigio_to_task(p, fown, fd, band);
488 } while_each_task_pid(-pid, PIDTYPE_PGID, p);
490 read_unlock(&tasklist_lock);
492 read_unlock(&fown->lock);
495 static void send_sigurg_to_task(struct task_struct *p,
496 struct fown_struct *fown)
498 if (sigio_perm(p, fown, SIGURG))
499 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
502 int send_sigurg(struct fown_struct *fown)
504 struct task_struct *p;
507 read_lock(&fown->lock);
510 goto out_unlock_fown;
514 read_lock(&tasklist_lock);
516 p = find_task_by_pid(pid);
518 send_sigurg_to_task(p, fown);
521 do_each_task_pid(-pid, PIDTYPE_PGID, p) {
522 send_sigurg_to_task(p, fown);
523 } while_each_task_pid(-pid, PIDTYPE_PGID, p);
525 read_unlock(&tasklist_lock);
527 read_unlock(&fown->lock);
531 static DEFINE_RWLOCK(fasync_lock);
532 static kmem_cache_t *fasync_cache;
535 * fasync_helper() is used by some character device drivers (mainly mice)
536 * to set up the fasync queue. It returns negative on error, 0 if it did
537 * no changes and positive if it added/deleted the entry.
539 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
541 struct fasync_struct *fa, **fp;
542 struct fasync_struct *new = NULL;
546 new = kmem_cache_alloc(fasync_cache, SLAB_KERNEL);
550 write_lock_irq(&fasync_lock);
551 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
552 if (fa->fa_file == filp) {
555 kmem_cache_free(fasync_cache, new);
558 kmem_cache_free(fasync_cache, fa);
566 new->magic = FASYNC_MAGIC;
569 new->fa_next = *fapp;
574 write_unlock_irq(&fasync_lock);
578 EXPORT_SYMBOL(fasync_helper);
580 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
583 struct fown_struct * fown;
584 if (fa->magic != FASYNC_MAGIC) {
585 printk(KERN_ERR "kill_fasync: bad magic number in "
589 fown = &fa->fa_file->f_owner;
590 /* Don't send SIGURG to processes which have not set a
591 queued signum: SIGURG has its own default signalling
593 if (!(sig == SIGURG && fown->signum == 0))
594 send_sigio(fown, fa->fa_fd, band);
599 EXPORT_SYMBOL(__kill_fasync);
601 void kill_fasync(struct fasync_struct **fp, int sig, int band)
603 /* First a quick test without locking: usually
607 read_lock(&fasync_lock);
608 /* reread *fp after obtaining the lock */
609 __kill_fasync(*fp, sig, band);
610 read_unlock(&fasync_lock);
613 EXPORT_SYMBOL(kill_fasync);
615 static int __init fasync_init(void)
617 fasync_cache = kmem_cache_create("fasync_cache",
618 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL, NULL);
622 module_init(fasync_init)